Take one task that keeps landing on you and write it down precisely enough to hand to somebody else. Or to something else.
Sort your week into four buckets and bucket three is the pile that repeats, has rules, and hurts you when you skip it.
That pile is not a to do list. It is a set of machines nobody has built yet.
Here is the part almost everyone gets wrong. The hard part is not the technology. It is that the task lives in your head as a vague sense of what should happen, and you cannot hand a vague feeling to anyone. Not to a new hire, and definitely not to a machine.
The rule page prompt below drags it out of your head and onto one page. The sort prompt comes first, for anyone who has not run the week yet.
Needs your judgment, your relationships, or your name on it.
Needs judgment, but not yours.
Repeats, has rules, and hurts you when you skip it. This page is about this one.
Habit. Nobody would notice if it stopped.
Start here if you have not sorted a week yet. This one reads your last seven days and puts every item into exactly one of the four buckets, with the reason in six words or less.
You are my operations advisor. You are blunt and you do not flatter me. Before you do any work, ask me these one at a time and wait for each answer: - Which calendar should you read, and on which account? - Where does my task list live? - Should you read my sent mail for the same seven days? Then pull my last seven days yourself. It will be messy. Do not ask me to tidy it. If you cannot reach something, name the missing connection and tell me what to connect. Never guess, and never work from an earlier session. Sort every item into exactly one of four buckets: 1. ONLY ME - needs my judgment, my relationships, or my name on it. 2. A PERSON'S JOB - needs judgment, but not mine. 3. A MACHINE'S JOB - repeats, has rules, and hurts me when I skip it. 4. SHOULD NOT EXIST - I do it out of habit and nobody would notice if it stopped. For each item, give the reason in six words or less. Then tell me: - The one item in bucket 3 costing me the most, and why you picked it. - The one item in bucket 4 you are most confident about, and what happens if I stop doing it. - One thing I put in bucket 1 that you think I am wrong about. End with one sentence: what my week says about what I actually believe my job is.
Pick the single item from bucket three that cost you the most this week. One item. Not the list.
Before you do any work, ask me these one at a time and wait for each answer: - Where does this task actually run today? Which tool, which inbox, which folder? - What can you reach of it right now, and what is missing? - Where should the finished rule page be saved? Then read what you can reach of how the task runs today. Do not ask me for anything you can look up yourself. I have one task that repeats, has rules, and costs me when I skip it. Help me write it down precisely enough that somebody else could make it run without me. The task is: [describe it in your own words, however badly] Ask me one question at a time until you can answer all of these. Wait for each answer before asking the next: - What starts it? A day, an event, a message, a threshold? - What information does it need, and where does that live today? - What does a good outcome look like, specifically? - What does a bad outcome look like, and how would I find out? - What is the one case where a human has to look at it? Then write THE RULE PAGE: TRIGGER - what starts it, stated so a stranger could recognise it. INPUTS - what it needs, and where each thing comes from. Name the actual tool or folder, not a category. THE RULE - what should happen, every time, no exceptions. Numbered. EXCEPTIONS - the cases that stop and wait for a person. Be strict. Most tasks have zero or one, not five. DONE - how anyone can tell it worked. FAILURE - what it should do when something is missing. Never guess, never invent, never proceed on partial information. Write it so a competent stranger could follow it with no access to me. Then tell me the one sentence on the page that is still too vague to hand to anybody, and ask me the question that would fix it. Last, ask me one question and wait for my answer: would I like help installing this machine? If I say yes, write me a short plain-English brief of what it must do, what it must never do, and how anyone would know it worked. Keep it under 150 words so I can hand it to somebody exactly as it stands.
The test of a finished rule page: hand it to someone who has never done the task and do not help them. If they finish without asking you a single question, you no longer own that task. If they come back with one question, that question is the sentence you still owe the page.
Got a brief you want built? Send it to leo@onetimeautomation.ai. Paste the brief exactly as the prompt wrote it. A real person reads every one, and if it is something you could install yourself in an afternoon, we will tell you that instead.
Recommended: Claude Code, Cowork, or Codex. These prompts work best in an AI agent that can reach your tools, rather than in a chat window. The sort prompt reads your calendar, task list, and sent mail. The rule page prompt reads wherever the task already runs, and writes the finished page back.